From 33ad03f085e54e61d3dc8a1e038a60f164ad890a Mon Sep 17 00:00:00 2001
From: jlqian98 <jlqian98@163.com>
Date: Thu, 19 Mar 2026 13:51:38 +0800
Subject: [PATCH] fix: resolve "negative dimensions are not allowed" on windows

---
 preprocess/tools/vocal_detection.py |  3 +++
 preprocess/utils.py                 | 18 +++++++++---------
 2 files changed, 12 insertions(+), 9 deletions(-)

diff --git a/preprocess/tools/vocal_detection.py b/preprocess/tools/vocal_detection.py
index 87bf499..05eac06 100644
--- a/preprocess/tools/vocal_detection.py
+++ b/preprocess/tools/vocal_detection.py
@@ -289,6 +289,9 @@ class VocalDetector:
             key = f"{base_name_no_ext}_{idx}"
             start_sample = librosa.time_to_samples(start_ms / 1000, sr=sr)
             end_sample = librosa.time_to_samples(end_ms / 1000, sr=sr)
+            # Use Python int to avoid 32-bit overflow on Windows
+            start_sample = int(start_sample)
+            end_sample = int(end_sample)
             segment = y[start_sample:end_sample]
 
             write(f"{self.cut_wavs_output_dir}/{key}.wav", segment, sr)
diff --git a/preprocess/utils.py b/preprocess/utils.py
index 745ace9..7ad2703 100644
--- a/preprocess/utils.py
+++ b/preprocess/utils.py
@@ -109,10 +109,11 @@ def _merge_group(
     )
 
     # ---------- time & audio ----------
-    start_ms = get_attr(segments[0], "start_time_ms", 0)
-    end_ms = get_attr(segments[-1], "end_time_ms", 0) + end_extension_ms
-    start_sample = start_ms * sample_rate // 1000
-    end_sample = end_ms * sample_rate // 1000
+    # Use Python int to avoid 32-bit overflow on Windows
+    start_ms = int(get_attr(segments[0], "start_time_ms", 0))
+    end_ms = int(get_attr(segments[-1], "end_time_ms", 0)) + end_extension_ms
+    start_sample = int(start_ms * sample_rate // 1000)
+    end_sample = int(end_ms * sample_rate // 1000)
 
     # ---------- naming ----------
     first_item_name = get_attr(segments[0], "item_name", "segment")
@@ -195,12 +196,11 @@ def merge_short_segments(
 
     for seg in segments:
         if isinstance(seg, dict):
-            start_time = seg.get("start_time_ms", 0)
-            end_time = seg.get("end_time_ms", 0)
+            start_time = int(seg.get("start_time_ms", 0))
+            end_time = int(seg.get("end_time_ms", 0))
         else:
-            start_time = seg.start_time_ms
-            end_time = seg.end_time_ms
-        
+            start_time = int(seg.start_time_ms)
+            end_time = int(seg.end_time_ms)
         if (
             current_group
             and (start_time - prev_end > max_gap_ms